Bridging Geographic Bias in Urban Streetscape Inference via Lifelong Learning with Visual-Semantic Pivoting

2026-06-12

AI Summary

This study addresses the systemic misjudgments and geographic biases in existing urban streetscape perception models, which stem from training data concentrated in a few major cities and adversely affect downstream decision-making. To mitigate these issues, the authors propose the HVSP-LL framework, which enables continual cross-city learning through a macro–meso–micro three-tier semantic anchoring system, integrated with a fairness-aware worst-region reweighting strategy and a structure-aware memory replay mechanism. The approach effectively alleviates distributional shift and catastrophic forgetting, achieving a Spearman correlation of 0.834 on a benchmark spanning twelve cities across four continents—improving by 6.1 points over the strongest baseline—and reduces inter-city perception disparity to 0.094, a reduction of over 38%, thereby substantially enhancing both model generalization and fairness.

Abstract

Visual perception of urban streetscapes underpins evidence-based decisions in landscape planning, public health, and place-making. Yet models trained on a few well-photographed metropolises systematically misjudge underrepresented districts, propagating geographic bias into downstream policy. We address this gap with HVSP-LL, a lifelong learning framework that couples a stratified visual-semantic pivoting module with an equity-aware rehearsal mechanism. The pivoting module organises landscape concepts along a three-tier ontology (macro structure, meso composition, micro element) and aligns image features to learnable semantic anchors at each tier, providing transferable representations that resist distributional drift. The lifelong adaptation component sequentially absorbs new urban regions while constraining inter-region perception gaps through a worst-region sample-reweighting objective and a structurally-aware exemplar buffer. We evaluate HVSP-LL on a panoramic streetscape benchmark assembled from twelve cities across four continents and seven perceptual dimensions. The framework attains 0.834 Spearman correlation on the held-out city sequence, an absolute 6.1 point improvement over the strongest continual baseline, and shrinks the inter-city perception gap to 0.094 -- a 38% reduction relative to the strongest continual baseline (0.151) and a 57% reduction relative to a representative regularisation baseline (0.218). Ablations confirm that each tier of the pivoting hierarchy contributes monotonically, and the equity-aware rehearsal converts mean backward transfer from -0.038 (without retention) to +0.013, eliminating catastrophic forgetting on the held-out sequence. Our results indicate that hierarchical anchoring is a practical pathway toward geographically equitable streetscape inference at city scale.

Problem

• geographic bias
• urban streetscape inference
• visual perception
• equity-aware learning
• distributional drift

Innovation

• lifelong learning
• visual-semantic pivoting
• geographic bias mitigation
• hierarchical ontology
• equity-aware rehearsal